File size: 8,448 Bytes
853865d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
{
 "name": "OpenGlyph-OCR-0",
 "released": "2026-11-14",
 "what": "one model in two forms. At the root, the form transformers loads with from_pretrained: model.safetensors, a checkpoint in the layout of PyTorch, with its configuration, tokenizer and model class. Beside it, the form for MLX: Xiaomi-OCR-0 unchanged (base/), the arrays trained on top of it (openglyph-ocr-0.safetensors) and the tokens added (tokens.json), which openglyph_ocr.py builds the model from",
 "base": {
  "model": "SeerRay-Lab/Xiaomi-OCR-0",
  "obtained_from": "ModelScope official model mirror",
  "repository_revision": "c0af34f6",
  "changed": false,
  "on_the_hugging_face_hub": {
   "repository": "xiaomi-research/Xiaomi-OCR-0",
   "revision": "bbd8cf8d21f256b3644b44cb860f3a7cecb93c30",
   "the_six_files_of_base_are_the_same_by_sha256": true,
   "checked_on": "2026-11-14"
  }
 },
 "trained_arrays": {
  "arrays": 148,
  "weights": 205304652,
  "vision_blocks_8_to_11_and_patch_merger": {
   "arrays": 54,
   "from_checkpoint": "step-0002048-final-1794217522541640000",
   "of_file_with_sha256": "5d61a67eba654bcc9cdceec17725de89704596f040262d6e746193d92720f7ae"
  },
  "lora_and_output_rows": {
   "arrays": 94,
   "from_checkpoint": "step-0011500-balanced-best-1794344748876820000",
   "of_file_with_sha256": "1c20cfa0004215544c3c2c8d3e74e315d180b30b97976cd481ebf0415653e337"
  },
  "held_bit_for_bit_as_the_checkpoints_hold_them": true
 },
 "added_tokens": {
  "first_id": 248077,
  "catalog_forms": 229,
  "extension_characters": 80992,
  "glyphwiki_glyphs": 77831,
  "vocabulary": 407129
 },
 "lora": {
  "rank": 8,
  "alpha": 1.0,
  "language_layers": [
   18,
   19,
   20,
   21,
   22,
   23
  ]
 },
 "transformers_form": {
  "file": "model.safetensors",
  "arrays": 490,
  "made_from": {
   "base/model.safetensors": "d4551e9f6127ac9e35cff7b4f7a72a90410e03e6c8676a36f0e2b7eaeb5d5348",
   "openglyph-ocr-0.safetensors": "69c7e631cb6bc85b4cd2451d1a28eedb4a0dd285ad836d31e88deefd82309f35"
  },
  "copied_from_the_base_byte_for_byte": 387,
  "stored_as_32_bit_floats": {
   "weights_with_the_lora_arrays_folded_in": 46,
   "retrained_vision_arrays": 54,
   "output_layer": [
    "lm_head.weight",
    "lm_head.bias"
   ]
  },
  "input_embedding": "the base's rows for the ids below 248077, then one and the same row for every added token (16-bit floats, as the base's)",
  "output_layer": "lm_head.weight: the base's rows, then the trained row of every added token; lm_head.bias: nought, then the trained bias of every added token; not tied to the input embedding",
  "every_array_read_back_against_its_source": true,
  "load_in": "float32",
  "tokenizer": {
   "added_tokens": 159052,
   "first_id": 248077,
   "vocabulary_and_merges": "the base's",
   "model_keys_that_differ": [],
   "differs_from_tokenizer_json_of_the_base_in": [
    "normalizer",
    "pre_tokenizer"
   ],
   "why": "it is the tokenizer of training, saved: transformers 5.18.0 builds the splitting of Qwen3.5 by itself (its pre_tokenizer), and training had the NFC normalisation switched off. tokenizer_config.json names the class TokenizersBackend so that this file is taken as it stands"
  }
 },
 "read_with_the_model_before_release": {
  "mlx_form": {
   "how": "the folder's own loader read lines of five tests again; each had to give the token ids and the text of the measured reading",
   "read_from": "the folder as it is uploaded",
   "on": "2026-11-14",
   "lines": 100,
   "read_as_measured": 100,
   "by_test": {
    "full_lines": {
     "lines": 20,
     "read_as_measured": 20,
     "seconds_a_line": 1.12
    },
    "hard_real_test": {
     "lines": 20,
     "read_as_measured": 20,
     "seconds_a_line": 0.49
    },
    "gw_test": {
     "lines": 20,
     "read_as_measured": 20,
     "seconds_a_line": 0.62
    },
    "ext_test": {
     "lines": 20,
     "read_as_measured": 20,
     "seconds_a_line": 0.66
    },
    "page_p08": {
     "lines": 20,
     "read_as_measured": 20,
     "seconds_a_line": 1.13
    }
   },
   "seconds_to_load": 7.0,
   "peak_gb": 3.94
  },
  "transformers_form": {
   "how": "loaded with AutoModelForImageTextToText.from_pretrained(trust_remote_code=True, dtype=float32); lines of the same tests read with read_line and held against the measured readings, which were made with the MLX form",
   "read_from": "the folder as it is uploaded",
   "on": "2026-11-14",
   "device": "cpu",
   "torch": "2.8.0",
   "transformers": "5.18.0",
   "number_types": [
    "torch.float32"
   ],
   "status": "stopped at its time limit; the lines read until then are counted",
   "lines": 43,
   "same_token_ids_as_measured": 39,
   "by_test": {
    "full_lines": {
     "lines": 9,
     "same_token_ids": 9
    },
    "hard_real_test": {
     "lines": 9,
     "same_token_ids": 8
    },
    "gw_test": {
     "lines": 9,
     "same_token_ids": 9
    },
    "ext_test": {
     "lines": 8,
     "same_token_ids": 5
    },
    "page_p08": {
     "lines": 8,
     "same_token_ids": 8
    }
   },
   "lines_that_differ": {
    "lines": 4,
    "the_measured_reading_is_the_answer": 0,
    "this_reading_is_the_answer": 0,
    "neither_is": 4,
    "both_are_through_other_tokens": 0
   },
   "seconds_a_line": 37.86,
   "seconds_to_load": 15.1
  }
 },
 "measured_with": {
  "python": "3.12.13",
  "platform": "macOS, Apple silicon",
  "mlx": "0.32.3",
  "mlx-vlm": "0.3.12",
  "mlx-lm": "0.32.0",
  "transformers": "5.18.0",
  "tokenizers": "0.23.2",
  "numpy": "2.5.3",
  "pillow": "12.3.0",
  "safetensors": "0.8.0",
  "torch": "2.8.0",
  "torchvision": "0.23.0"
 },
 "files": {
  "LICENSE": {
   "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
   "bytes": 11358
  },
  "NOTICE": {
   "sha256": "d5b6636468e146160e7a28d6e6055e9f781cf25f7575cf215e86c094a4c8f501",
   "bytes": 2162
  },
  "README.md": {
   "sha256": "7fee5b471f1f3001a8fea621d8728c35dc2f37432dc31dbb7b1588bc0a8ba67d",
   "bytes": 15439
  },
  "base/chat_template.jinja": {
   "sha256": "273d8e0e683b885071fb17e08d71e5f2a5ddfb5309756181681de4f5a1822d80",
   "bytes": 7755
  },
  "base/config.json": {
   "sha256": "2eb95982069a24336fe66b3a0869ea5a711aba7566787d4a599654907ecc070c",
   "bytes": 2747
  },
  "base/model.safetensors": {
   "sha256": "d4551e9f6127ac9e35cff7b4f7a72a90410e03e6c8676a36f0e2b7eaeb5d5348",
   "bytes": 1746937856
  },
  "base/processor_config.json": {
   "sha256": "d89ef49ce9cd37fbf510158e13c1ef063d9286411c1ec9049932dbe0487143b1",
   "bytes": 1191
  },
  "base/tokenizer.json": {
   "sha256": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523",
   "bytes": 19989325
  },
  "base/tokenizer_config.json": {
   "sha256": "792fa3f0cb88b111e54ef3134c873531008c4df471d108da17903426e308aa7b",
   "bytes": 1165
  },
  "chat_template.jinja": {
   "sha256": "273d8e0e683b885071fb17e08d71e5f2a5ddfb5309756181681de4f5a1822d80",
   "bytes": 7755
  },
  "config.json": {
   "sha256": "ce369170e6b1c42923230356b3c1a87ae49f9c0dd0eba719f4271ee5b0b768f9",
   "bytes": 3083
  },
  "configuration_openglyph_ocr.py": {
   "sha256": "f11da637f204313f177a54e099987aef159d6eec1ce9334c6103f73284c02760",
   "bytes": 617
  },
  "generation_config.json": {
   "sha256": "c5a3c8614e67c379a09a10522d43bc0a6137bbbf6c32c31528415b9ea92e5b3e",
   "bytes": 102
  },
  "model.safetensors": {
   "sha256": "b158b9e6b8a6ca236aa648e6ee32d30185764e9a3954a00bb87f531363887b49",
   "bytes": 4068916884
  },
  "modeling_openglyph_ocr.py": {
   "sha256": "2e892ef75f6d85ce897500d17fd864fa685bd5842b517e03be4c00aaca95e707",
   "bytes": 4453
  },
  "openglyph-ocr-0.safetensors": {
   "sha256": "69c7e631cb6bc85b4cd2451d1a28eedb4a0dd285ad836d31e88deefd82309f35",
   "bytes": 821236272
  },
  "openglyph_ocr.py": {
   "sha256": "a597edd13bab6304cb37364456f4630c0e789247bce9dbf357f9029f2af8c768",
   "bytes": 13782
  },
  "processor_config.json": {
   "sha256": "bfbc24af59a3e73a9cd0653b8d4ae758dfaec4e3e6c15dfb1c6c8ee8d5683c85",
   "bytes": 1220
  },
  "requirements.txt": {
   "sha256": "8b649edd2bfb0dbe93ce8a4176143eb941cf820e6b570d06e151f63c19c5e85d",
   "bytes": 576
  },
  "tokenizer.json": {
   "sha256": "6b0afa98d4b6b1803326ffbbd0af210fbee81aeb9fee18ef83d6626f3ebd27a1",
   "bytes": 50448682
  },
  "tokenizer_config.json": {
   "sha256": "69dcb12d80d243f24e7808985b772b62625fdb8fd92644b59c1e679c6c1e81e8",
   "bytes": 1191
  },
  "tokens.json": {
   "sha256": "50165ec65b8a5a358b684cd84519ff6487eea3b8902cffac01843aebe3248368",
   "bytes": 2784718
  }
 }
}